]>
Commit | Line | Data |
---|---|---|
893f7eba PB |
1 | /* |
2 | * Image mirroring | |
3 | * | |
4 | * Copyright Red Hat, Inc. 2012 | |
5 | * | |
6 | * Authors: | |
7 | * Paolo Bonzini <pbonzini@redhat.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU LGPL, version 2 or later. | |
10 | * See the COPYING.LIB file in the top-level directory. | |
11 | * | |
12 | */ | |
13 | ||
80c71a24 | 14 | #include "qemu/osdep.h" |
fd4a6493 | 15 | #include "qemu/cutils.h" |
893f7eba | 16 | #include "trace.h" |
c87621ea | 17 | #include "block/blockjob_int.h" |
737e150e | 18 | #include "block/block_int.h" |
373340b2 | 19 | #include "sysemu/block-backend.h" |
da34e65c | 20 | #include "qapi/error.h" |
cc7a8ea7 | 21 | #include "qapi/qmp/qerror.h" |
893f7eba | 22 | #include "qemu/ratelimit.h" |
b812f671 | 23 | #include "qemu/bitmap.h" |
893f7eba | 24 | |
402a4741 PB |
25 | #define SLICE_TIME 100000000ULL /* ns */ |
26 | #define MAX_IN_FLIGHT 16 | |
b436982f EB |
27 | #define MAX_IO_BYTES (1 << 20) /* 1 Mb */ |
28 | #define DEFAULT_MIRROR_BUF_SIZE (MAX_IN_FLIGHT * MAX_IO_BYTES) | |
402a4741 PB |
29 | |
30 | /* The mirroring buffer is a list of granularity-sized chunks. | |
31 | * Free chunks are organized in a list. | |
32 | */ | |
33 | typedef struct MirrorBuffer { | |
34 | QSIMPLEQ_ENTRY(MirrorBuffer) next; | |
35 | } MirrorBuffer; | |
893f7eba PB |
36 | |
37 | typedef struct MirrorBlockJob { | |
38 | BlockJob common; | |
39 | RateLimit limit; | |
e253f4b8 | 40 | BlockBackend *target; |
4ef85a9c KW |
41 | BlockDriverState *mirror_top_bs; |
42 | BlockDriverState *source; | |
5bc361b8 | 43 | BlockDriverState *base; |
4ef85a9c | 44 | |
09158f00 BC |
45 | /* The name of the graph node to replace */ |
46 | char *replaces; | |
47 | /* The BDS to replace */ | |
48 | BlockDriverState *to_replace; | |
49 | /* Used to block operations on the drive-mirror-replace target */ | |
50 | Error *replace_blocker; | |
03544a6e | 51 | bool is_none_mode; |
274fccee | 52 | BlockMirrorBackingMode backing_mode; |
b952b558 | 53 | BlockdevOnError on_source_error, on_target_error; |
d63ffd87 PB |
54 | bool synced; |
55 | bool should_complete; | |
eee13dfe | 56 | int64_t granularity; |
b812f671 | 57 | size_t buf_size; |
b21c7652 | 58 | int64_t bdev_length; |
b812f671 | 59 | unsigned long *cow_bitmap; |
e4654d2d | 60 | BdrvDirtyBitmap *dirty_bitmap; |
dc162c8e | 61 | BdrvDirtyBitmapIter *dbi; |
893f7eba | 62 | uint8_t *buf; |
402a4741 PB |
63 | QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; |
64 | int buf_free_count; | |
bd48bde8 | 65 | |
49efb1f5 | 66 | uint64_t last_pause_ns; |
402a4741 | 67 | unsigned long *in_flight_bitmap; |
bd48bde8 | 68 | int in_flight; |
b436982f | 69 | int64_t bytes_in_flight; |
bd48bde8 | 70 | int ret; |
0fc9f8ea | 71 | bool unmap; |
e424aff5 | 72 | bool waiting_for_io; |
b436982f | 73 | int target_cluster_size; |
e5b43573 | 74 | int max_iov; |
90ab48eb | 75 | bool initial_zeroing_ongoing; |
893f7eba PB |
76 | } MirrorBlockJob; |
77 | ||
bd48bde8 PB |
78 | typedef struct MirrorOp { |
79 | MirrorBlockJob *s; | |
80 | QEMUIOVector qiov; | |
b436982f EB |
81 | int64_t offset; |
82 | uint64_t bytes; | |
bd48bde8 PB |
83 | } MirrorOp; |
84 | ||
b952b558 PB |
85 | static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, |
86 | int error) | |
87 | { | |
88 | s->synced = false; | |
89 | if (read) { | |
81e254dc KW |
90 | return block_job_error_action(&s->common, s->on_source_error, |
91 | true, error); | |
b952b558 | 92 | } else { |
81e254dc KW |
93 | return block_job_error_action(&s->common, s->on_target_error, |
94 | false, error); | |
b952b558 PB |
95 | } |
96 | } | |
97 | ||
bd48bde8 PB |
98 | static void mirror_iteration_done(MirrorOp *op, int ret) |
99 | { | |
100 | MirrorBlockJob *s = op->s; | |
402a4741 | 101 | struct iovec *iov; |
bd48bde8 | 102 | int64_t chunk_num; |
b436982f | 103 | int i, nb_chunks; |
bd48bde8 | 104 | |
b436982f | 105 | trace_mirror_iteration_done(s, op->offset, op->bytes, ret); |
bd48bde8 PB |
106 | |
107 | s->in_flight--; | |
b436982f | 108 | s->bytes_in_flight -= op->bytes; |
402a4741 PB |
109 | iov = op->qiov.iov; |
110 | for (i = 0; i < op->qiov.niov; i++) { | |
111 | MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; | |
112 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); | |
113 | s->buf_free_count++; | |
114 | } | |
115 | ||
b436982f EB |
116 | chunk_num = op->offset / s->granularity; |
117 | nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); | |
402a4741 | 118 | bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); |
b21c7652 HR |
119 | if (ret >= 0) { |
120 | if (s->cow_bitmap) { | |
121 | bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); | |
122 | } | |
90ab48eb | 123 | if (!s->initial_zeroing_ongoing) { |
b436982f | 124 | s->common.offset += op->bytes; |
90ab48eb | 125 | } |
bd48bde8 | 126 | } |
6df3bf8e | 127 | qemu_iovec_destroy(&op->qiov); |
c84b3192 | 128 | g_free(op); |
7b770c72 | 129 | |
e424aff5 | 130 | if (s->waiting_for_io) { |
0b8b8753 | 131 | qemu_coroutine_enter(s->common.co); |
7b770c72 | 132 | } |
bd48bde8 PB |
133 | } |
134 | ||
135 | static void mirror_write_complete(void *opaque, int ret) | |
136 | { | |
137 | MirrorOp *op = opaque; | |
138 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
139 | |
140 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 141 | if (ret < 0) { |
bd48bde8 PB |
142 | BlockErrorAction action; |
143 | ||
b436982f EB |
144 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset >> BDRV_SECTOR_BITS, |
145 | op->bytes >> BDRV_SECTOR_BITS); | |
bd48bde8 | 146 | action = mirror_error_action(s, false, -ret); |
a589569f | 147 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
148 | s->ret = ret; |
149 | } | |
150 | } | |
151 | mirror_iteration_done(op, ret); | |
b9e413dd | 152 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
153 | } |
154 | ||
155 | static void mirror_read_complete(void *opaque, int ret) | |
156 | { | |
157 | MirrorOp *op = opaque; | |
158 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
159 | |
160 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 161 | if (ret < 0) { |
bd48bde8 PB |
162 | BlockErrorAction action; |
163 | ||
b436982f EB |
164 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset >> BDRV_SECTOR_BITS, |
165 | op->bytes >> BDRV_SECTOR_BITS); | |
bd48bde8 | 166 | action = mirror_error_action(s, true, -ret); |
a589569f | 167 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
168 | s->ret = ret; |
169 | } | |
170 | ||
171 | mirror_iteration_done(op, ret); | |
b9e413dd | 172 | } else { |
b436982f | 173 | blk_aio_pwritev(s->target, op->offset, &op->qiov, |
b9e413dd | 174 | 0, mirror_write_complete, op); |
bd48bde8 | 175 | } |
b9e413dd | 176 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
177 | } |
178 | ||
782d97ef EB |
179 | /* Clip bytes relative to offset to not exceed end-of-file */ |
180 | static inline int64_t mirror_clip_bytes(MirrorBlockJob *s, | |
181 | int64_t offset, | |
182 | int64_t bytes) | |
183 | { | |
184 | return MIN(bytes, s->bdev_length - offset); | |
185 | } | |
186 | ||
187 | /* Clip nb_sectors relative to sector_num to not exceed end-of-file */ | |
931e5260 EB |
188 | static inline int mirror_clip_sectors(MirrorBlockJob *s, |
189 | int64_t sector_num, | |
190 | int nb_sectors) | |
4150ae60 | 191 | { |
931e5260 EB |
192 | return MIN(nb_sectors, |
193 | s->bdev_length / BDRV_SECTOR_SIZE - sector_num); | |
4150ae60 FZ |
194 | } |
195 | ||
782d97ef EB |
196 | /* Round offset and/or bytes to target cluster if COW is needed, and |
197 | * return the offset of the adjusted tail against original. */ | |
198 | static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset, | |
ae4cc877 | 199 | uint64_t *bytes) |
893f7eba | 200 | { |
e5b43573 FZ |
201 | bool need_cow; |
202 | int ret = 0; | |
782d97ef EB |
203 | int64_t align_offset = *offset; |
204 | unsigned int align_bytes = *bytes; | |
205 | int max_bytes = s->granularity * s->max_iov; | |
e5b43573 | 206 | |
ae4cc877 | 207 | assert(*bytes < INT_MAX); |
782d97ef EB |
208 | need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap); |
209 | need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity, | |
e5b43573 FZ |
210 | s->cow_bitmap); |
211 | if (need_cow) { | |
782d97ef EB |
212 | bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes, |
213 | &align_offset, &align_bytes); | |
e5b43573 | 214 | } |
3515727f | 215 | |
782d97ef EB |
216 | if (align_bytes > max_bytes) { |
217 | align_bytes = max_bytes; | |
e5b43573 | 218 | if (need_cow) { |
782d97ef | 219 | align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size); |
e5b43573 | 220 | } |
8f0720ec | 221 | } |
782d97ef | 222 | /* Clipping may result in align_bytes unaligned to chunk boundary, but |
4150ae60 | 223 | * that doesn't matter because it's already the end of source image. */ |
782d97ef | 224 | align_bytes = mirror_clip_bytes(s, align_offset, align_bytes); |
8f0720ec | 225 | |
782d97ef EB |
226 | ret = align_offset + align_bytes - (*offset + *bytes); |
227 | *offset = align_offset; | |
228 | *bytes = align_bytes; | |
e5b43573 FZ |
229 | assert(ret >= 0); |
230 | return ret; | |
231 | } | |
232 | ||
21cd917f FZ |
233 | static inline void mirror_wait_for_io(MirrorBlockJob *s) |
234 | { | |
235 | assert(!s->waiting_for_io); | |
236 | s->waiting_for_io = true; | |
237 | qemu_coroutine_yield(); | |
238 | s->waiting_for_io = false; | |
239 | } | |
240 | ||
e5b43573 | 241 | /* Submit async read while handling COW. |
ae4cc877 EB |
242 | * Returns: The number of bytes copied after and including offset, |
243 | * excluding any bytes copied prior to offset due to alignment. | |
244 | * This will be @bytes if no alignment is necessary, or | |
245 | * (new_end - offset) if tail is rounded up or down due to | |
e5b43573 FZ |
246 | * alignment or buffer limit. |
247 | */ | |
ae4cc877 EB |
248 | static uint64_t mirror_do_read(MirrorBlockJob *s, int64_t offset, |
249 | uint64_t bytes) | |
e5b43573 | 250 | { |
e253f4b8 | 251 | BlockBackend *source = s->common.blk; |
ae4cc877 EB |
252 | int nb_chunks; |
253 | uint64_t ret; | |
e5b43573 | 254 | MirrorOp *op; |
ae4cc877 | 255 | uint64_t max_bytes; |
e5b43573 | 256 | |
ae4cc877 | 257 | max_bytes = s->granularity * s->max_iov; |
402a4741 | 258 | |
e5b43573 | 259 | /* We can only handle as much as buf_size at a time. */ |
ae4cc877 EB |
260 | bytes = MIN(s->buf_size, MIN(max_bytes, bytes)); |
261 | assert(bytes); | |
262 | assert(bytes < BDRV_REQUEST_MAX_BYTES); | |
263 | ret = bytes; | |
402a4741 | 264 | |
e5b43573 | 265 | if (s->cow_bitmap) { |
ae4cc877 | 266 | ret += mirror_cow_align(s, &offset, &bytes); |
e5b43573 | 267 | } |
ae4cc877 EB |
268 | assert(bytes <= s->buf_size); |
269 | /* The offset is granularity-aligned because: | |
e5b43573 FZ |
270 | * 1) Caller passes in aligned values; |
271 | * 2) mirror_cow_align is used only when target cluster is larger. */ | |
ae4cc877 EB |
272 | assert(QEMU_IS_ALIGNED(offset, s->granularity)); |
273 | /* The range is sector-aligned, since bdrv_getlength() rounds up. */ | |
274 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
275 | nb_chunks = DIV_ROUND_UP(bytes, s->granularity); | |
e5b43573 FZ |
276 | |
277 | while (s->buf_free_count < nb_chunks) { | |
ae4cc877 | 278 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
21cd917f | 279 | mirror_wait_for_io(s); |
b812f671 PB |
280 | } |
281 | ||
bd48bde8 | 282 | /* Allocate a MirrorOp that is used as an AIO callback. */ |
c84b3192 | 283 | op = g_new(MirrorOp, 1); |
bd48bde8 | 284 | op->s = s; |
ae4cc877 EB |
285 | op->offset = offset; |
286 | op->bytes = bytes; | |
402a4741 PB |
287 | |
288 | /* Now make a QEMUIOVector taking enough granularity-sized chunks | |
289 | * from s->buf_free. | |
290 | */ | |
291 | qemu_iovec_init(&op->qiov, nb_chunks); | |
402a4741 PB |
292 | while (nb_chunks-- > 0) { |
293 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
ae4cc877 | 294 | size_t remaining = bytes - op->qiov.size; |
5a0f6fd5 | 295 | |
402a4741 PB |
296 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
297 | s->buf_free_count--; | |
5a0f6fd5 | 298 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 | 299 | } |
bd48bde8 | 300 | |
893f7eba | 301 | /* Copy the dirty cluster. */ |
bd48bde8 | 302 | s->in_flight++; |
ae4cc877 EB |
303 | s->bytes_in_flight += bytes; |
304 | trace_mirror_one_iteration(s, offset, bytes); | |
dcfb3beb | 305 | |
ae4cc877 | 306 | blk_aio_preadv(source, offset, &op->qiov, 0, mirror_read_complete, op); |
e5b43573 FZ |
307 | return ret; |
308 | } | |
309 | ||
310 | static void mirror_do_zero_or_discard(MirrorBlockJob *s, | |
e6f24193 EB |
311 | int64_t offset, |
312 | uint64_t bytes, | |
e5b43573 FZ |
313 | bool is_discard) |
314 | { | |
315 | MirrorOp *op; | |
316 | ||
317 | /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed | |
318 | * so the freeing in mirror_iteration_done is nop. */ | |
319 | op = g_new0(MirrorOp, 1); | |
320 | op->s = s; | |
e6f24193 EB |
321 | op->offset = offset; |
322 | op->bytes = bytes; | |
e5b43573 FZ |
323 | |
324 | s->in_flight++; | |
e6f24193 | 325 | s->bytes_in_flight += bytes; |
e5b43573 | 326 | if (is_discard) { |
e6f24193 | 327 | blk_aio_pdiscard(s->target, offset, |
b436982f | 328 | op->bytes, mirror_write_complete, op); |
e5b43573 | 329 | } else { |
e6f24193 | 330 | blk_aio_pwrite_zeroes(s->target, offset, |
b436982f | 331 | op->bytes, s->unmap ? BDRV_REQ_MAY_UNMAP : 0, |
dcfb3beb | 332 | mirror_write_complete, op); |
e5b43573 FZ |
333 | } |
334 | } | |
335 | ||
336 | static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) | |
337 | { | |
4ef85a9c | 338 | BlockDriverState *source = s->source; |
9c83625b | 339 | int64_t sector_num, first_chunk; |
e5b43573 FZ |
340 | uint64_t delay_ns = 0; |
341 | /* At least the first dirty chunk is mirrored in one iteration. */ | |
342 | int nb_chunks = 1; | |
343 | int64_t end = s->bdev_length / BDRV_SECTOR_SIZE; | |
344 | int sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; | |
4b5004d9 | 345 | bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); |
b436982f | 346 | int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES); |
e5b43573 | 347 | |
b64bd51e | 348 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
dc162c8e | 349 | sector_num = bdrv_dirty_iter_next(s->dbi); |
e5b43573 | 350 | if (sector_num < 0) { |
dc162c8e FZ |
351 | bdrv_set_dirty_iter(s->dbi, 0); |
352 | sector_num = bdrv_dirty_iter_next(s->dbi); | |
5cb1a49e EB |
353 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap) * |
354 | BDRV_SECTOR_SIZE); | |
e5b43573 FZ |
355 | assert(sector_num >= 0); |
356 | } | |
b64bd51e | 357 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
e5b43573 | 358 | |
9c83625b HR |
359 | first_chunk = sector_num / sectors_per_chunk; |
360 | while (test_bit(first_chunk, s->in_flight_bitmap)) { | |
5cb1a49e EB |
361 | trace_mirror_yield_in_flight(s, sector_num * BDRV_SECTOR_SIZE, |
362 | s->in_flight); | |
9c83625b HR |
363 | mirror_wait_for_io(s); |
364 | } | |
365 | ||
565ac01f SH |
366 | block_job_pause_point(&s->common); |
367 | ||
e5b43573 FZ |
368 | /* Find the number of consective dirty chunks following the first dirty |
369 | * one, and wait for in flight requests in them. */ | |
b64bd51e | 370 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
e5b43573 | 371 | while (nb_chunks * sectors_per_chunk < (s->buf_size >> BDRV_SECTOR_BITS)) { |
dc162c8e | 372 | int64_t next_dirty; |
e5b43573 FZ |
373 | int64_t next_sector = sector_num + nb_chunks * sectors_per_chunk; |
374 | int64_t next_chunk = next_sector / sectors_per_chunk; | |
375 | if (next_sector >= end || | |
b64bd51e | 376 | !bdrv_get_dirty_locked(source, s->dirty_bitmap, next_sector)) { |
e5b43573 FZ |
377 | break; |
378 | } | |
379 | if (test_bit(next_chunk, s->in_flight_bitmap)) { | |
9c83625b | 380 | break; |
e5b43573 | 381 | } |
9c83625b | 382 | |
dc162c8e FZ |
383 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
384 | if (next_dirty > next_sector || next_dirty < 0) { | |
f27a2742 | 385 | /* The bitmap iterator's cache is stale, refresh it */ |
dc162c8e FZ |
386 | bdrv_set_dirty_iter(s->dbi, next_sector); |
387 | next_dirty = bdrv_dirty_iter_next(s->dbi); | |
f27a2742 | 388 | } |
dc162c8e | 389 | assert(next_dirty == next_sector); |
9c83625b | 390 | nb_chunks++; |
e5b43573 FZ |
391 | } |
392 | ||
393 | /* Clear dirty bits before querying the block status, because | |
394 | * calling bdrv_get_block_status_above could yield - if some blocks are | |
395 | * marked dirty in this window, we need to know. | |
396 | */ | |
b64bd51e PB |
397 | bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, sector_num, |
398 | nb_chunks * sectors_per_chunk); | |
399 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); | |
400 | ||
e5b43573 FZ |
401 | bitmap_set(s->in_flight_bitmap, sector_num / sectors_per_chunk, nb_chunks); |
402 | while (nb_chunks > 0 && sector_num < end) { | |
39c11580 | 403 | int64_t ret; |
f3e4ce4a EB |
404 | int io_sectors; |
405 | int64_t io_bytes_acct; | |
e5b43573 FZ |
406 | BlockDriverState *file; |
407 | enum MirrorMethod { | |
408 | MIRROR_METHOD_COPY, | |
409 | MIRROR_METHOD_ZERO, | |
410 | MIRROR_METHOD_DISCARD | |
411 | } mirror_method = MIRROR_METHOD_COPY; | |
412 | ||
413 | assert(!(sector_num % sectors_per_chunk)); | |
414 | ret = bdrv_get_block_status_above(source, NULL, sector_num, | |
415 | nb_chunks * sectors_per_chunk, | |
416 | &io_sectors, &file); | |
417 | if (ret < 0) { | |
b436982f EB |
418 | io_sectors = MIN(nb_chunks * sectors_per_chunk, |
419 | max_io_bytes >> BDRV_SECTOR_BITS); | |
0965a41e | 420 | } else if (ret & BDRV_BLOCK_DATA) { |
b436982f | 421 | io_sectors = MIN(io_sectors, max_io_bytes >> BDRV_SECTOR_BITS); |
e5b43573 FZ |
422 | } |
423 | ||
424 | io_sectors -= io_sectors % sectors_per_chunk; | |
425 | if (io_sectors < sectors_per_chunk) { | |
426 | io_sectors = sectors_per_chunk; | |
427 | } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { | |
428 | int64_t target_sector_num; | |
429 | int target_nb_sectors; | |
244483e6 KW |
430 | bdrv_round_sectors_to_clusters(blk_bs(s->target), sector_num, |
431 | io_sectors, &target_sector_num, | |
432 | &target_nb_sectors); | |
e5b43573 FZ |
433 | if (target_sector_num == sector_num && |
434 | target_nb_sectors == io_sectors) { | |
435 | mirror_method = ret & BDRV_BLOCK_ZERO ? | |
436 | MIRROR_METHOD_ZERO : | |
437 | MIRROR_METHOD_DISCARD; | |
438 | } | |
439 | } | |
440 | ||
cf56a3c6 | 441 | while (s->in_flight >= MAX_IN_FLIGHT) { |
5cb1a49e EB |
442 | trace_mirror_yield_in_flight(s, sector_num * BDRV_SECTOR_SIZE, |
443 | s->in_flight); | |
cf56a3c6 DL |
444 | mirror_wait_for_io(s); |
445 | } | |
446 | ||
dbaa7b57 VSO |
447 | if (s->ret < 0) { |
448 | return 0; | |
449 | } | |
450 | ||
931e5260 | 451 | io_sectors = mirror_clip_sectors(s, sector_num, io_sectors); |
e5b43573 FZ |
452 | switch (mirror_method) { |
453 | case MIRROR_METHOD_COPY: | |
ae4cc877 EB |
454 | io_bytes_acct = mirror_do_read(s, sector_num * BDRV_SECTOR_SIZE, |
455 | io_sectors * BDRV_SECTOR_SIZE); | |
456 | io_sectors = io_bytes_acct / BDRV_SECTOR_SIZE; | |
e5b43573 FZ |
457 | break; |
458 | case MIRROR_METHOD_ZERO: | |
e5b43573 | 459 | case MIRROR_METHOD_DISCARD: |
e6f24193 EB |
460 | mirror_do_zero_or_discard(s, sector_num * BDRV_SECTOR_SIZE, |
461 | io_sectors * BDRV_SECTOR_SIZE, | |
4b5004d9 DL |
462 | mirror_method == MIRROR_METHOD_DISCARD); |
463 | if (write_zeroes_ok) { | |
f3e4ce4a | 464 | io_bytes_acct = 0; |
4b5004d9 | 465 | } else { |
f3e4ce4a | 466 | io_bytes_acct = io_sectors * BDRV_SECTOR_SIZE; |
4b5004d9 | 467 | } |
e5b43573 FZ |
468 | break; |
469 | default: | |
470 | abort(); | |
471 | } | |
472 | assert(io_sectors); | |
473 | sector_num += io_sectors; | |
4150ae60 | 474 | nb_chunks -= DIV_ROUND_UP(io_sectors, sectors_per_chunk); |
f14a39cc | 475 | if (s->common.speed) { |
f3e4ce4a | 476 | delay_ns = ratelimit_calculate_delay(&s->limit, io_bytes_acct); |
f14a39cc | 477 | } |
dcfb3beb | 478 | } |
cc8c9d6c | 479 | return delay_ns; |
bd48bde8 | 480 | } |
b952b558 | 481 | |
402a4741 PB |
482 | static void mirror_free_init(MirrorBlockJob *s) |
483 | { | |
484 | int granularity = s->granularity; | |
485 | size_t buf_size = s->buf_size; | |
486 | uint8_t *buf = s->buf; | |
487 | ||
488 | assert(s->buf_free_count == 0); | |
489 | QSIMPLEQ_INIT(&s->buf_free); | |
490 | while (buf_size != 0) { | |
491 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
492 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
493 | s->buf_free_count++; | |
494 | buf_size -= granularity; | |
495 | buf += granularity; | |
496 | } | |
497 | } | |
498 | ||
bae8196d PB |
499 | /* This is also used for the .pause callback. There is no matching |
500 | * mirror_resume() because mirror_run() will begin iterating again | |
501 | * when the job is resumed. | |
502 | */ | |
503 | static void mirror_wait_for_all_io(MirrorBlockJob *s) | |
bd48bde8 PB |
504 | { |
505 | while (s->in_flight > 0) { | |
21cd917f | 506 | mirror_wait_for_io(s); |
bd48bde8 | 507 | } |
893f7eba PB |
508 | } |
509 | ||
5a7e7a0b SH |
510 | typedef struct { |
511 | int ret; | |
512 | } MirrorExitData; | |
513 | ||
514 | static void mirror_exit(BlockJob *job, void *opaque) | |
515 | { | |
516 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
517 | MirrorExitData *data = opaque; | |
518 | AioContext *replace_aio_context = NULL; | |
4ef85a9c | 519 | BlockDriverState *src = s->source; |
e253f4b8 | 520 | BlockDriverState *target_bs = blk_bs(s->target); |
4ef85a9c | 521 | BlockDriverState *mirror_top_bs = s->mirror_top_bs; |
12fa4af6 | 522 | Error *local_err = NULL; |
3f09bfbc | 523 | |
2119882c PB |
524 | bdrv_release_dirty_bitmap(src, s->dirty_bitmap); |
525 | ||
3f09bfbc KW |
526 | /* Make sure that the source BDS doesn't go away before we called |
527 | * block_job_completed(). */ | |
528 | bdrv_ref(src); | |
4ef85a9c | 529 | bdrv_ref(mirror_top_bs); |
7d9fcb39 KW |
530 | bdrv_ref(target_bs); |
531 | ||
532 | /* Remove target parent that still uses BLK_PERM_WRITE/RESIZE before | |
533 | * inserting target_bs at s->to_replace, where we might not be able to get | |
63c8ef28 KW |
534 | * these permissions. |
535 | * | |
536 | * Note that blk_unref() alone doesn't necessarily drop permissions because | |
537 | * we might be running nested inside mirror_drain(), which takes an extra | |
538 | * reference, so use an explicit blk_set_perm() first. */ | |
539 | blk_set_perm(s->target, 0, BLK_PERM_ALL, &error_abort); | |
7d9fcb39 KW |
540 | blk_unref(s->target); |
541 | s->target = NULL; | |
4ef85a9c KW |
542 | |
543 | /* We don't access the source any more. Dropping any WRITE/RESIZE is | |
544 | * required before it could become a backing file of target_bs. */ | |
545 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, | |
546 | &error_abort); | |
547 | if (s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { | |
548 | BlockDriverState *backing = s->is_none_mode ? src : s->base; | |
549 | if (backing_bs(target_bs) != backing) { | |
12fa4af6 KW |
550 | bdrv_set_backing_hd(target_bs, backing, &local_err); |
551 | if (local_err) { | |
552 | error_report_err(local_err); | |
553 | data->ret = -EPERM; | |
554 | } | |
4ef85a9c KW |
555 | } |
556 | } | |
5a7e7a0b SH |
557 | |
558 | if (s->to_replace) { | |
559 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
560 | aio_context_acquire(replace_aio_context); | |
561 | } | |
562 | ||
563 | if (s->should_complete && data->ret == 0) { | |
e253f4b8 | 564 | BlockDriverState *to_replace = src; |
5a7e7a0b SH |
565 | if (s->to_replace) { |
566 | to_replace = s->to_replace; | |
567 | } | |
40365552 | 568 | |
e253f4b8 KW |
569 | if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) { |
570 | bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL); | |
5a7e7a0b | 571 | } |
b8804815 KW |
572 | |
573 | /* The mirror job has no requests in flight any more, but we need to | |
574 | * drain potential other users of the BDS before changing the graph. */ | |
e253f4b8 | 575 | bdrv_drained_begin(target_bs); |
5fe31c25 | 576 | bdrv_replace_node(to_replace, target_bs, &local_err); |
e253f4b8 | 577 | bdrv_drained_end(target_bs); |
5fe31c25 KW |
578 | if (local_err) { |
579 | error_report_err(local_err); | |
580 | data->ret = -EPERM; | |
581 | } | |
5a7e7a0b SH |
582 | } |
583 | if (s->to_replace) { | |
584 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
585 | error_free(s->replace_blocker); | |
586 | bdrv_unref(s->to_replace); | |
587 | } | |
588 | if (replace_aio_context) { | |
589 | aio_context_release(replace_aio_context); | |
590 | } | |
591 | g_free(s->replaces); | |
7d9fcb39 | 592 | bdrv_unref(target_bs); |
4ef85a9c KW |
593 | |
594 | /* Remove the mirror filter driver from the graph. Before this, get rid of | |
595 | * the blockers on the intermediate nodes so that the resulting state is | |
0bf74767 KW |
596 | * valid. Also give up permissions on mirror_top_bs->backing, which might |
597 | * block the removal. */ | |
4ef85a9c | 598 | block_job_remove_all_bdrv(job); |
c1cef672 FZ |
599 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
600 | &error_abort); | |
5fe31c25 | 601 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
4ef85a9c KW |
602 | |
603 | /* We just changed the BDS the job BB refers to (with either or both of the | |
5fe31c25 KW |
604 | * bdrv_replace_node() calls), so switch the BB back so the cleanup does |
605 | * the right thing. We don't need any permissions any more now. */ | |
4ef85a9c KW |
606 | blk_remove_bs(job->blk); |
607 | blk_set_perm(job->blk, 0, BLK_PERM_ALL, &error_abort); | |
608 | blk_insert_bs(job->blk, mirror_top_bs, &error_abort); | |
609 | ||
5a7e7a0b | 610 | block_job_completed(&s->common, data->ret); |
4ef85a9c | 611 | |
5a7e7a0b | 612 | g_free(data); |
176c3699 | 613 | bdrv_drained_end(src); |
4ef85a9c | 614 | bdrv_unref(mirror_top_bs); |
3f09bfbc | 615 | bdrv_unref(src); |
5a7e7a0b SH |
616 | } |
617 | ||
49efb1f5 DL |
618 | static void mirror_throttle(MirrorBlockJob *s) |
619 | { | |
620 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
621 | ||
622 | if (now - s->last_pause_ns > SLICE_TIME) { | |
623 | s->last_pause_ns = now; | |
624 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0); | |
625 | } else { | |
626 | block_job_pause_point(&s->common); | |
627 | } | |
628 | } | |
629 | ||
c0b363ad DL |
630 | static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) |
631 | { | |
632 | int64_t sector_num, end; | |
633 | BlockDriverState *base = s->base; | |
4ef85a9c | 634 | BlockDriverState *bs = s->source; |
c0b363ad | 635 | BlockDriverState *target_bs = blk_bs(s->target); |
c0b363ad DL |
636 | int ret, n; |
637 | ||
638 | end = s->bdev_length / BDRV_SECTOR_SIZE; | |
639 | ||
b7d5062c | 640 | if (base == NULL && !bdrv_has_zero_init(target_bs)) { |
c7c2769c DL |
641 | if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { |
642 | bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, end); | |
643 | return 0; | |
644 | } | |
645 | ||
90ab48eb | 646 | s->initial_zeroing_ongoing = true; |
c7c2769c DL |
647 | for (sector_num = 0; sector_num < end; ) { |
648 | int nb_sectors = MIN(end - sector_num, | |
649 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity) >> BDRV_SECTOR_BITS); | |
650 | ||
651 | mirror_throttle(s); | |
652 | ||
653 | if (block_job_is_cancelled(&s->common)) { | |
90ab48eb | 654 | s->initial_zeroing_ongoing = false; |
c7c2769c DL |
655 | return 0; |
656 | } | |
657 | ||
658 | if (s->in_flight >= MAX_IN_FLIGHT) { | |
67adf4b3 EB |
659 | trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, |
660 | s->in_flight); | |
c7c2769c DL |
661 | mirror_wait_for_io(s); |
662 | continue; | |
663 | } | |
664 | ||
e6f24193 EB |
665 | mirror_do_zero_or_discard(s, sector_num * BDRV_SECTOR_SIZE, |
666 | nb_sectors * BDRV_SECTOR_SIZE, false); | |
c7c2769c DL |
667 | sector_num += nb_sectors; |
668 | } | |
669 | ||
bae8196d | 670 | mirror_wait_for_all_io(s); |
90ab48eb | 671 | s->initial_zeroing_ongoing = false; |
b7d5062c DL |
672 | } |
673 | ||
c0b363ad DL |
674 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
675 | for (sector_num = 0; sector_num < end; ) { | |
676 | /* Just to make sure we are not exceeding int limit. */ | |
677 | int nb_sectors = MIN(INT_MAX >> BDRV_SECTOR_BITS, | |
678 | end - sector_num); | |
679 | ||
680 | mirror_throttle(s); | |
681 | ||
682 | if (block_job_is_cancelled(&s->common)) { | |
683 | return 0; | |
684 | } | |
685 | ||
686 | ret = bdrv_is_allocated_above(bs, base, sector_num, nb_sectors, &n); | |
687 | if (ret < 0) { | |
688 | return ret; | |
689 | } | |
690 | ||
691 | assert(n > 0); | |
b7d5062c | 692 | if (ret == 1) { |
c0b363ad DL |
693 | bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n); |
694 | } | |
695 | sector_num += n; | |
696 | } | |
697 | return 0; | |
698 | } | |
699 | ||
bdffb31d PB |
700 | /* Called when going out of the streaming phase to flush the bulk of the |
701 | * data to the medium, or just before completing. | |
702 | */ | |
703 | static int mirror_flush(MirrorBlockJob *s) | |
704 | { | |
705 | int ret = blk_flush(s->target); | |
706 | if (ret < 0) { | |
707 | if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { | |
708 | s->ret = ret; | |
709 | } | |
710 | } | |
711 | return ret; | |
712 | } | |
713 | ||
893f7eba PB |
714 | static void coroutine_fn mirror_run(void *opaque) |
715 | { | |
716 | MirrorBlockJob *s = opaque; | |
5a7e7a0b | 717 | MirrorExitData *data; |
4ef85a9c | 718 | BlockDriverState *bs = s->source; |
e253f4b8 | 719 | BlockDriverState *target_bs = blk_bs(s->target); |
9a0cec66 | 720 | bool need_drain = true; |
c0b363ad | 721 | int64_t length; |
b812f671 | 722 | BlockDriverInfo bdi; |
1d33936e JC |
723 | char backing_filename[2]; /* we only need 2 characters because we are only |
724 | checking for a NULL string */ | |
893f7eba | 725 | int ret = 0; |
893f7eba PB |
726 | |
727 | if (block_job_is_cancelled(&s->common)) { | |
728 | goto immediate_exit; | |
729 | } | |
730 | ||
b21c7652 HR |
731 | s->bdev_length = bdrv_getlength(bs); |
732 | if (s->bdev_length < 0) { | |
733 | ret = s->bdev_length; | |
373df5b1 | 734 | goto immediate_exit; |
becc347e KW |
735 | } |
736 | ||
737 | /* Active commit must resize the base image if its size differs from the | |
738 | * active layer. */ | |
739 | if (s->base == blk_bs(s->target)) { | |
740 | int64_t base_length; | |
741 | ||
742 | base_length = blk_getlength(s->target); | |
743 | if (base_length < 0) { | |
744 | ret = base_length; | |
745 | goto immediate_exit; | |
746 | } | |
747 | ||
748 | if (s->bdev_length > base_length) { | |
ed3d2ec9 | 749 | ret = blk_truncate(s->target, s->bdev_length, NULL); |
becc347e KW |
750 | if (ret < 0) { |
751 | goto immediate_exit; | |
752 | } | |
753 | } | |
754 | } | |
755 | ||
756 | if (s->bdev_length == 0) { | |
9e48b025 FZ |
757 | /* Report BLOCK_JOB_READY and wait for complete. */ |
758 | block_job_event_ready(&s->common); | |
759 | s->synced = true; | |
760 | while (!block_job_is_cancelled(&s->common) && !s->should_complete) { | |
761 | block_job_yield(&s->common); | |
762 | } | |
763 | s->common.cancelled = false; | |
764 | goto immediate_exit; | |
893f7eba PB |
765 | } |
766 | ||
b21c7652 | 767 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
768 | s->in_flight_bitmap = bitmap_new(length); |
769 | ||
b812f671 PB |
770 | /* If we have no backing file yet in the destination, we cannot let |
771 | * the destination do COW. Instead, we copy sectors around the | |
772 | * dirty data if needed. We need a bitmap to do that. | |
773 | */ | |
e253f4b8 | 774 | bdrv_get_backing_filename(target_bs, backing_filename, |
b812f671 | 775 | sizeof(backing_filename)); |
e253f4b8 | 776 | if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { |
b436982f EB |
777 | s->target_cluster_size = bdi.cluster_size; |
778 | } else { | |
779 | s->target_cluster_size = BDRV_SECTOR_SIZE; | |
e5b43573 | 780 | } |
b436982f EB |
781 | if (backing_filename[0] && !target_bs->backing && |
782 | s->granularity < s->target_cluster_size) { | |
783 | s->buf_size = MAX(s->buf_size, s->target_cluster_size); | |
e5b43573 | 784 | s->cow_bitmap = bitmap_new(length); |
b812f671 | 785 | } |
e253f4b8 | 786 | s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); |
b812f671 | 787 | |
7504edf4 KW |
788 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
789 | if (s->buf == NULL) { | |
790 | ret = -ENOMEM; | |
791 | goto immediate_exit; | |
792 | } | |
793 | ||
402a4741 | 794 | mirror_free_init(s); |
893f7eba | 795 | |
49efb1f5 | 796 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 797 | if (!s->is_none_mode) { |
c0b363ad DL |
798 | ret = mirror_dirty_init(s); |
799 | if (ret < 0 || block_job_is_cancelled(&s->common)) { | |
800 | goto immediate_exit; | |
893f7eba PB |
801 | } |
802 | } | |
803 | ||
dc162c8e FZ |
804 | assert(!s->dbi); |
805 | s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap, 0); | |
893f7eba | 806 | for (;;) { |
cc8c9d6c | 807 | uint64_t delay_ns = 0; |
49efb1f5 | 808 | int64_t cnt, delta; |
893f7eba PB |
809 | bool should_complete; |
810 | ||
bd48bde8 PB |
811 | if (s->ret < 0) { |
812 | ret = s->ret; | |
813 | goto immediate_exit; | |
814 | } | |
815 | ||
565ac01f SH |
816 | block_job_pause_point(&s->common); |
817 | ||
20dca810 | 818 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
b21c7652 HR |
819 | /* s->common.offset contains the number of bytes already processed so |
820 | * far, cnt is the number of dirty sectors remaining and | |
b436982f | 821 | * s->bytes_in_flight is the number of bytes currently being |
b21c7652 | 822 | * processed; together those are the current total operation length */ |
b436982f EB |
823 | s->common.len = s->common.offset + s->bytes_in_flight + |
824 | cnt * BDRV_SECTOR_SIZE; | |
bd48bde8 PB |
825 | |
826 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 827 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
bd48bde8 PB |
828 | * We do so every SLICE_TIME nanoseconds, or when there is an error, |
829 | * or when the source is clean, whichever comes first. | |
830 | */ | |
49efb1f5 DL |
831 | delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; |
832 | if (delta < SLICE_TIME && | |
bd48bde8 | 833 | s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
cf56a3c6 | 834 | if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || |
402a4741 | 835 | (cnt == 0 && s->in_flight > 0)) { |
5cb1a49e EB |
836 | trace_mirror_yield(s, cnt * BDRV_SECTOR_SIZE, |
837 | s->buf_free_count, s->in_flight); | |
21cd917f | 838 | mirror_wait_for_io(s); |
bd48bde8 PB |
839 | continue; |
840 | } else if (cnt != 0) { | |
cc8c9d6c | 841 | delay_ns = mirror_iteration(s); |
893f7eba | 842 | } |
893f7eba PB |
843 | } |
844 | ||
845 | should_complete = false; | |
bd48bde8 | 846 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba | 847 | trace_mirror_before_flush(s); |
bdffb31d PB |
848 | if (!s->synced) { |
849 | if (mirror_flush(s) < 0) { | |
850 | /* Go check s->ret. */ | |
851 | continue; | |
b952b558 | 852 | } |
b952b558 PB |
853 | /* We're out of the streaming phase. From now on, if the job |
854 | * is cancelled we will actually complete all pending I/O and | |
855 | * report completion. This way, block-job-cancel will leave | |
856 | * the target in a consistent state. | |
857 | */ | |
bdffb31d PB |
858 | block_job_event_ready(&s->common); |
859 | s->synced = true; | |
d63ffd87 | 860 | } |
bdffb31d PB |
861 | |
862 | should_complete = s->should_complete || | |
863 | block_job_is_cancelled(&s->common); | |
864 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); | |
893f7eba PB |
865 | } |
866 | ||
867 | if (cnt == 0 && should_complete) { | |
868 | /* The dirty bitmap is not updated while operations are pending. | |
869 | * If we're about to exit, wait for pending operations before | |
870 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
871 | * source has dirty data to copy! | |
872 | * | |
873 | * Note that I/O can be submitted by the guest while | |
9a0cec66 PB |
874 | * mirror_populate runs, so pause it now. Before deciding |
875 | * whether to switch to target check one last time if I/O has | |
876 | * come in the meanwhile, and if not flush the data to disk. | |
893f7eba | 877 | */ |
5cb1a49e | 878 | trace_mirror_before_drain(s, cnt * BDRV_SECTOR_SIZE); |
9a0cec66 PB |
879 | |
880 | bdrv_drained_begin(bs); | |
20dca810 | 881 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
bdffb31d | 882 | if (cnt > 0 || mirror_flush(s) < 0) { |
9a0cec66 PB |
883 | bdrv_drained_end(bs); |
884 | continue; | |
885 | } | |
886 | ||
887 | /* The two disks are in sync. Exit and report successful | |
888 | * completion. | |
889 | */ | |
890 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
891 | s->common.cancelled = false; | |
892 | need_drain = false; | |
893 | break; | |
893f7eba PB |
894 | } |
895 | ||
896 | ret = 0; | |
5cb1a49e EB |
897 | trace_mirror_before_sleep(s, cnt * BDRV_SECTOR_SIZE, |
898 | s->synced, delay_ns); | |
d63ffd87 | 899 | if (!s->synced) { |
7483d1e5 | 900 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba PB |
901 | if (block_job_is_cancelled(&s->common)) { |
902 | break; | |
903 | } | |
904 | } else if (!should_complete) { | |
bd48bde8 | 905 | delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); |
7483d1e5 | 906 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba | 907 | } |
49efb1f5 | 908 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
909 | } |
910 | ||
911 | immediate_exit: | |
bd48bde8 PB |
912 | if (s->in_flight > 0) { |
913 | /* We get here only if something went wrong. Either the job failed, | |
914 | * or it was cancelled prematurely so that we do not guarantee that | |
915 | * the target is a copy of the source. | |
916 | */ | |
917 | assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common))); | |
9a0cec66 | 918 | assert(need_drain); |
bae8196d | 919 | mirror_wait_for_all_io(s); |
bd48bde8 PB |
920 | } |
921 | ||
922 | assert(s->in_flight == 0); | |
7191bf31 | 923 | qemu_vfree(s->buf); |
b812f671 | 924 | g_free(s->cow_bitmap); |
402a4741 | 925 | g_free(s->in_flight_bitmap); |
dc162c8e | 926 | bdrv_dirty_iter_free(s->dbi); |
5a7e7a0b SH |
927 | |
928 | data = g_malloc(sizeof(*data)); | |
929 | data->ret = ret; | |
9a0cec66 PB |
930 | |
931 | if (need_drain) { | |
932 | bdrv_drained_begin(bs); | |
933 | } | |
5a7e7a0b | 934 | block_job_defer_to_main_loop(&s->common, mirror_exit, data); |
893f7eba PB |
935 | } |
936 | ||
937 | static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) | |
938 | { | |
939 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
940 | ||
941 | if (speed < 0) { | |
c6bd8c70 | 942 | error_setg(errp, QERR_INVALID_PARAMETER, "speed"); |
893f7eba PB |
943 | return; |
944 | } | |
f3e4ce4a | 945 | ratelimit_set_speed(&s->limit, speed, SLICE_TIME); |
893f7eba PB |
946 | } |
947 | ||
d63ffd87 PB |
948 | static void mirror_complete(BlockJob *job, Error **errp) |
949 | { | |
950 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
4ef85a9c | 951 | BlockDriverState *target; |
274fccee | 952 | |
274fccee | 953 | target = blk_bs(s->target); |
d63ffd87 | 954 | |
d63ffd87 | 955 | if (!s->synced) { |
9df229c3 AG |
956 | error_setg(errp, "The active block job '%s' cannot be completed", |
957 | job->id); | |
d63ffd87 PB |
958 | return; |
959 | } | |
960 | ||
274fccee HR |
961 | if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { |
962 | int ret; | |
963 | ||
964 | assert(!target->backing); | |
965 | ret = bdrv_open_backing_file(target, NULL, "backing", errp); | |
966 | if (ret < 0) { | |
967 | return; | |
968 | } | |
969 | } | |
970 | ||
15d67298 | 971 | /* block all operations on to_replace bs */ |
09158f00 | 972 | if (s->replaces) { |
5a7e7a0b SH |
973 | AioContext *replace_aio_context; |
974 | ||
e12f3784 | 975 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 976 | if (!s->to_replace) { |
e12f3784 | 977 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
978 | return; |
979 | } | |
980 | ||
5a7e7a0b SH |
981 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
982 | aio_context_acquire(replace_aio_context); | |
983 | ||
4ef85a9c KW |
984 | /* TODO Translate this into permission system. Current definition of |
985 | * GRAPH_MOD would require to request it for the parents; they might | |
986 | * not even be BlockDriverStates, however, so a BdrvChild can't address | |
987 | * them. May need redefinition of GRAPH_MOD. */ | |
09158f00 BC |
988 | error_setg(&s->replace_blocker, |
989 | "block device is in use by block-job-complete"); | |
990 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
991 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
992 | |
993 | aio_context_release(replace_aio_context); | |
09158f00 BC |
994 | } |
995 | ||
d63ffd87 | 996 | s->should_complete = true; |
751ebd76 | 997 | block_job_enter(&s->common); |
d63ffd87 PB |
998 | } |
999 | ||
bae8196d | 1000 | static void mirror_pause(BlockJob *job) |
565ac01f SH |
1001 | { |
1002 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
1003 | ||
bae8196d | 1004 | mirror_wait_for_all_io(s); |
565ac01f SH |
1005 | } |
1006 | ||
1007 | static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) | |
1008 | { | |
1009 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
1010 | ||
1011 | blk_set_aio_context(s->target, new_context); | |
1012 | } | |
1013 | ||
bae8196d PB |
1014 | static void mirror_drain(BlockJob *job) |
1015 | { | |
1016 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
1017 | ||
1018 | /* Need to keep a reference in case blk_drain triggers execution | |
1019 | * of mirror_complete... | |
1020 | */ | |
1021 | if (s->target) { | |
1022 | BlockBackend *target = s->target; | |
1023 | blk_ref(target); | |
1024 | blk_drain(target); | |
1025 | blk_unref(target); | |
1026 | } | |
1027 | } | |
1028 | ||
3fc4b10a | 1029 | static const BlockJobDriver mirror_job_driver = { |
565ac01f SH |
1030 | .instance_size = sizeof(MirrorBlockJob), |
1031 | .job_type = BLOCK_JOB_TYPE_MIRROR, | |
1032 | .set_speed = mirror_set_speed, | |
a7815a76 | 1033 | .start = mirror_run, |
565ac01f SH |
1034 | .complete = mirror_complete, |
1035 | .pause = mirror_pause, | |
1036 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 1037 | .drain = mirror_drain, |
893f7eba PB |
1038 | }; |
1039 | ||
03544a6e | 1040 | static const BlockJobDriver commit_active_job_driver = { |
565ac01f SH |
1041 | .instance_size = sizeof(MirrorBlockJob), |
1042 | .job_type = BLOCK_JOB_TYPE_COMMIT, | |
1043 | .set_speed = mirror_set_speed, | |
a7815a76 | 1044 | .start = mirror_run, |
565ac01f SH |
1045 | .complete = mirror_complete, |
1046 | .pause = mirror_pause, | |
1047 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 1048 | .drain = mirror_drain, |
03544a6e FZ |
1049 | }; |
1050 | ||
4ef85a9c KW |
1051 | static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, |
1052 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1053 | { | |
1054 | return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); | |
1055 | } | |
1056 | ||
1057 | static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, | |
1058 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1059 | { | |
1060 | return bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); | |
1061 | } | |
1062 | ||
1063 | static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) | |
1064 | { | |
1065 | return bdrv_co_flush(bs->backing->bs); | |
1066 | } | |
1067 | ||
1068 | static int64_t coroutine_fn bdrv_mirror_top_get_block_status( | |
1069 | BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum, | |
1070 | BlockDriverState **file) | |
1071 | { | |
1072 | *pnum = nb_sectors; | |
1073 | *file = bs->backing->bs; | |
d5254033 | 1074 | return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID | |
4ef85a9c KW |
1075 | (sector_num << BDRV_SECTOR_BITS); |
1076 | } | |
1077 | ||
1078 | static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, | |
f5a5ca79 | 1079 | int64_t offset, int bytes, BdrvRequestFlags flags) |
4ef85a9c | 1080 | { |
f5a5ca79 | 1081 | return bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); |
4ef85a9c KW |
1082 | } |
1083 | ||
1084 | static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, | |
f5a5ca79 | 1085 | int64_t offset, int bytes) |
4ef85a9c | 1086 | { |
f5a5ca79 | 1087 | return bdrv_co_pdiscard(bs->backing->bs, offset, bytes); |
4ef85a9c KW |
1088 | } |
1089 | ||
fd4a6493 KW |
1090 | static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs, QDict *opts) |
1091 | { | |
1092 | bdrv_refresh_filename(bs->backing->bs); | |
1093 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
1094 | bs->backing->bs->filename); | |
1095 | } | |
1096 | ||
4ef85a9c KW |
1097 | static void bdrv_mirror_top_close(BlockDriverState *bs) |
1098 | { | |
1099 | } | |
1100 | ||
1101 | static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, | |
1102 | const BdrvChildRole *role, | |
1103 | uint64_t perm, uint64_t shared, | |
1104 | uint64_t *nperm, uint64_t *nshared) | |
1105 | { | |
1106 | /* Must be able to forward guest writes to the real image */ | |
1107 | *nperm = 0; | |
1108 | if (perm & BLK_PERM_WRITE) { | |
1109 | *nperm |= BLK_PERM_WRITE; | |
1110 | } | |
1111 | ||
1112 | *nshared = BLK_PERM_ALL; | |
1113 | } | |
1114 | ||
1115 | /* Dummy node that provides consistent read to its users without requiring it | |
1116 | * from its backing file and that allows writes on the backing file chain. */ | |
1117 | static BlockDriver bdrv_mirror_top = { | |
1118 | .format_name = "mirror_top", | |
1119 | .bdrv_co_preadv = bdrv_mirror_top_preadv, | |
1120 | .bdrv_co_pwritev = bdrv_mirror_top_pwritev, | |
1121 | .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, | |
1122 | .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, | |
1123 | .bdrv_co_flush = bdrv_mirror_top_flush, | |
1124 | .bdrv_co_get_block_status = bdrv_mirror_top_get_block_status, | |
fd4a6493 | 1125 | .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, |
4ef85a9c KW |
1126 | .bdrv_close = bdrv_mirror_top_close, |
1127 | .bdrv_child_perm = bdrv_mirror_top_child_perm, | |
1128 | }; | |
1129 | ||
71aa9867 | 1130 | static void mirror_start_job(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1131 | int creation_flags, BlockDriverState *target, |
1132 | const char *replaces, int64_t speed, | |
1133 | uint32_t granularity, int64_t buf_size, | |
274fccee | 1134 | BlockMirrorBackingMode backing_mode, |
09158f00 BC |
1135 | BlockdevOnError on_source_error, |
1136 | BlockdevOnError on_target_error, | |
0fc9f8ea | 1137 | bool unmap, |
097310b5 | 1138 | BlockCompletionFunc *cb, |
51ccfa2d | 1139 | void *opaque, |
09158f00 | 1140 | const BlockJobDriver *driver, |
b49f7ead | 1141 | bool is_none_mode, BlockDriverState *base, |
51ccfa2d FZ |
1142 | bool auto_complete, const char *filter_node_name, |
1143 | Error **errp) | |
893f7eba PB |
1144 | { |
1145 | MirrorBlockJob *s; | |
4ef85a9c KW |
1146 | BlockDriverState *mirror_top_bs; |
1147 | bool target_graph_mod; | |
1148 | bool target_is_backing; | |
b2c2832c | 1149 | Error *local_err = NULL; |
d7086422 | 1150 | int ret; |
893f7eba | 1151 | |
eee13dfe | 1152 | if (granularity == 0) { |
341ebc2f | 1153 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
1154 | } |
1155 | ||
1156 | assert ((granularity & (granularity - 1)) == 0); | |
b436982f EB |
1157 | /* Granularity must be large enough for sector-based dirty bitmap */ |
1158 | assert(granularity >= BDRV_SECTOR_SIZE); | |
eee13dfe | 1159 | |
48ac0a4d WC |
1160 | if (buf_size < 0) { |
1161 | error_setg(errp, "Invalid parameter 'buf-size'"); | |
1162 | return; | |
1163 | } | |
1164 | ||
1165 | if (buf_size == 0) { | |
1166 | buf_size = DEFAULT_MIRROR_BUF_SIZE; | |
1167 | } | |
5bc361b8 | 1168 | |
4ef85a9c KW |
1169 | /* In the case of active commit, add dummy driver to provide consistent |
1170 | * reads on the top, while disabling it in the intermediate nodes, and make | |
1171 | * the backing chain writable. */ | |
6cdbceb1 KW |
1172 | mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, |
1173 | BDRV_O_RDWR, errp); | |
4ef85a9c KW |
1174 | if (mirror_top_bs == NULL) { |
1175 | return; | |
1176 | } | |
1177 | mirror_top_bs->total_sectors = bs->total_sectors; | |
19dd29e8 | 1178 | bdrv_set_aio_context(mirror_top_bs, bdrv_get_aio_context(bs)); |
4ef85a9c KW |
1179 | |
1180 | /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep | |
7a25fcd0 | 1181 | * it alive until block_job_create() succeeds even if bs has no parent. */ |
4ef85a9c KW |
1182 | bdrv_ref(mirror_top_bs); |
1183 | bdrv_drained_begin(bs); | |
b2c2832c | 1184 | bdrv_append(mirror_top_bs, bs, &local_err); |
4ef85a9c KW |
1185 | bdrv_drained_end(bs); |
1186 | ||
b2c2832c KW |
1187 | if (local_err) { |
1188 | bdrv_unref(mirror_top_bs); | |
1189 | error_propagate(errp, local_err); | |
1190 | return; | |
1191 | } | |
1192 | ||
4ef85a9c KW |
1193 | /* Make sure that the source is not resized while the job is running */ |
1194 | s = block_job_create(job_id, driver, mirror_top_bs, | |
1195 | BLK_PERM_CONSISTENT_READ, | |
1196 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | | |
1197 | BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, | |
c6cc12bf | 1198 | creation_flags, cb, opaque, errp); |
893f7eba | 1199 | if (!s) { |
4ef85a9c | 1200 | goto fail; |
893f7eba | 1201 | } |
7a25fcd0 HR |
1202 | /* The block job now has a reference to this node */ |
1203 | bdrv_unref(mirror_top_bs); | |
1204 | ||
4ef85a9c KW |
1205 | s->source = bs; |
1206 | s->mirror_top_bs = mirror_top_bs; | |
1207 | ||
1208 | /* No resize for the target either; while the mirror is still running, a | |
1209 | * consistent read isn't necessarily possible. We could possibly allow | |
1210 | * writes and graph modifications, though it would likely defeat the | |
1211 | * purpose of a mirror, so leave them blocked for now. | |
1212 | * | |
1213 | * In the case of active commit, things look a bit different, though, | |
1214 | * because the target is an already populated backing file in active use. | |
1215 | * We can allow anything except resize there.*/ | |
1216 | target_is_backing = bdrv_chain_contains(bs, target); | |
1217 | target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN); | |
1218 | s->target = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE | | |
1219 | (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0), | |
1220 | BLK_PERM_WRITE_UNCHANGED | | |
1221 | (target_is_backing ? BLK_PERM_CONSISTENT_READ | | |
1222 | BLK_PERM_WRITE | | |
1223 | BLK_PERM_GRAPH_MOD : 0)); | |
d7086422 KW |
1224 | ret = blk_insert_bs(s->target, target, errp); |
1225 | if (ret < 0) { | |
4ef85a9c | 1226 | goto fail; |
d7086422 | 1227 | } |
e253f4b8 | 1228 | |
09158f00 | 1229 | s->replaces = g_strdup(replaces); |
b952b558 PB |
1230 | s->on_source_error = on_source_error; |
1231 | s->on_target_error = on_target_error; | |
03544a6e | 1232 | s->is_none_mode = is_none_mode; |
274fccee | 1233 | s->backing_mode = backing_mode; |
5bc361b8 | 1234 | s->base = base; |
eee13dfe | 1235 | s->granularity = granularity; |
48ac0a4d | 1236 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 1237 | s->unmap = unmap; |
b49f7ead WC |
1238 | if (auto_complete) { |
1239 | s->should_complete = true; | |
1240 | } | |
b812f671 | 1241 | |
0db6e54a | 1242 | s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); |
b8afb520 | 1243 | if (!s->dirty_bitmap) { |
88f9d1b3 | 1244 | goto fail; |
b8afb520 | 1245 | } |
10f3cd15 | 1246 | |
4ef85a9c | 1247 | /* Required permissions are already taken with blk_new() */ |
76d554e2 KW |
1248 | block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, |
1249 | &error_abort); | |
1250 | ||
f3ede4b0 AG |
1251 | /* In commit_active_start() all intermediate nodes disappear, so |
1252 | * any jobs in them must be blocked */ | |
4ef85a9c | 1253 | if (target_is_backing) { |
f3ede4b0 AG |
1254 | BlockDriverState *iter; |
1255 | for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { | |
4ef85a9c KW |
1256 | /* XXX BLK_PERM_WRITE needs to be allowed so we don't block |
1257 | * ourselves at s->base (if writes are blocked for a node, they are | |
1258 | * also blocked for its backing file). The other options would be a | |
1259 | * second filter driver above s->base (== target). */ | |
1260 | ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, | |
1261 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE, | |
1262 | errp); | |
1263 | if (ret < 0) { | |
1264 | goto fail; | |
1265 | } | |
f3ede4b0 AG |
1266 | } |
1267 | } | |
10f3cd15 | 1268 | |
5ccac6f1 JS |
1269 | trace_mirror_start(bs, s, opaque); |
1270 | block_job_start(&s->common); | |
4ef85a9c KW |
1271 | return; |
1272 | ||
1273 | fail: | |
1274 | if (s) { | |
7a25fcd0 HR |
1275 | /* Make sure this BDS does not go away until we have completed the graph |
1276 | * changes below */ | |
1277 | bdrv_ref(mirror_top_bs); | |
1278 | ||
4ef85a9c KW |
1279 | g_free(s->replaces); |
1280 | blk_unref(s->target); | |
05b0d8e3 | 1281 | block_job_early_fail(&s->common); |
4ef85a9c KW |
1282 | } |
1283 | ||
c1cef672 FZ |
1284 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
1285 | &error_abort); | |
5fe31c25 | 1286 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
7a25fcd0 HR |
1287 | |
1288 | bdrv_unref(mirror_top_bs); | |
893f7eba | 1289 | } |
03544a6e | 1290 | |
71aa9867 AG |
1291 | void mirror_start(const char *job_id, BlockDriverState *bs, |
1292 | BlockDriverState *target, const char *replaces, | |
5fba6c0e | 1293 | int64_t speed, uint32_t granularity, int64_t buf_size, |
274fccee HR |
1294 | MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, |
1295 | BlockdevOnError on_source_error, | |
03544a6e | 1296 | BlockdevOnError on_target_error, |
6cdbceb1 | 1297 | bool unmap, const char *filter_node_name, Error **errp) |
03544a6e FZ |
1298 | { |
1299 | bool is_none_mode; | |
1300 | BlockDriverState *base; | |
1301 | ||
4b80ab2b JS |
1302 | if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { |
1303 | error_setg(errp, "Sync mode 'incremental' not supported"); | |
d58d8453 JS |
1304 | return; |
1305 | } | |
03544a6e | 1306 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
760e0063 | 1307 | base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; |
47970dfb | 1308 | mirror_start_job(job_id, bs, BLOCK_JOB_DEFAULT, target, replaces, |
274fccee | 1309 | speed, granularity, buf_size, backing_mode, |
51ccfa2d | 1310 | on_source_error, on_target_error, unmap, NULL, NULL, |
6cdbceb1 | 1311 | &mirror_job_driver, is_none_mode, base, false, |
51ccfa2d | 1312 | filter_node_name, errp); |
03544a6e FZ |
1313 | } |
1314 | ||
fd62c609 | 1315 | void commit_active_start(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1316 | BlockDriverState *base, int creation_flags, |
1317 | int64_t speed, BlockdevOnError on_error, | |
0db832f4 | 1318 | const char *filter_node_name, |
78bbd910 FZ |
1319 | BlockCompletionFunc *cb, void *opaque, |
1320 | bool auto_complete, Error **errp) | |
03544a6e | 1321 | { |
4da83585 | 1322 | int orig_base_flags; |
cc67f4d1 | 1323 | Error *local_err = NULL; |
4da83585 JC |
1324 | |
1325 | orig_base_flags = bdrv_get_flags(base); | |
1326 | ||
20a63d2c FZ |
1327 | if (bdrv_reopen(base, bs->open_flags, errp)) { |
1328 | return; | |
1329 | } | |
4da83585 | 1330 | |
47970dfb | 1331 | mirror_start_job(job_id, bs, creation_flags, base, NULL, speed, 0, 0, |
71aa9867 | 1332 | MIRROR_LEAVE_BACKING_CHAIN, |
51ccfa2d | 1333 | on_error, on_error, true, cb, opaque, |
6cdbceb1 | 1334 | &commit_active_job_driver, false, base, auto_complete, |
51ccfa2d | 1335 | filter_node_name, &local_err); |
0fb6395c | 1336 | if (local_err) { |
cc67f4d1 | 1337 | error_propagate(errp, local_err); |
4da83585 JC |
1338 | goto error_restore_flags; |
1339 | } | |
1340 | ||
1341 | return; | |
1342 | ||
1343 | error_restore_flags: | |
1344 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
1345 | * the original error */ | |
1346 | bdrv_reopen(base, orig_base_flags, NULL); | |
1347 | return; | |
03544a6e | 1348 | } |